muapi-brochures
Pass
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-controlled inputs such as 'brand_name' and 'topic' and interpolates them directly into image generation prompts without sanitization or boundary markers to prevent potential instruction override.
- Ingestion points: User input fields defined in SKILL.md (e.g., brand_name, topic, style).
- Boundary markers: No delimiters or explicit instructions are provided to the agent to treat user data as untrusted.
- Capability inventory: The skill utilizes the muapi CLI and shell commands like curl for network operations.
- Sanitization: No evidence of validation, escaping, or filtering of the external user content.
- [COMMAND_EXECUTION]: The instructions direct the executing agent to run shell commands to interact with the Muapi service.
- Evidence: SKILL.md instructs the agent to execute 'muapi auth configure', 'muapi predict wait', and curl commands for API interaction.
- [DATA_EXFILTRATION]: The skill instructions specify transmitting the MUAPI_API_KEY environment variable to an external service.
- Evidence: SKILL.md includes a fallback instruction to use curl to send the MUAPI_API_KEY to api.muapi.ai via HTTP headers.
Audit Metadata